Skip to content
You are reading gnark development version documentation and some displayed features may not be available in the stable release. You can switch to stable version using the version box at screen bottom.

Updated on March 30, 2022


What’s gnark?

gnark is a fast zk-SNARK library that offers a high-level API to design circuits. The library is open source and developed under the Apache 2.0 license

How does gnark work?

In a typical workflow:

  1. Implement an algorithm for which you want to prove and verify execution.
  2. Use the gnark/frontend package to translate the algorithm into a set of mathematical constraints.
  3. Use the gnark/backend package to create and verify your proof of knowledge. That is, you prove that you know a list of secret inputs satisfying a set of mathematical constraints.


gnark has not been audited and is provided as-is, use at your own risk. In particular, gnark makes no security guarantees such as constant time implementation or side-channel attack resistance.

gnark circuits are written in Go

Users write their zk-SNARK circuits in plain Go. gnark uses Go because:

  • Go is a mature and widely used language with a robust tool chain.
  • Developers can debug, document, test and benchmark circuits as they would with any other Go program.
  • Circuits can be versioned, unit-tested and used in standard continuous integration and delivery (CI/CD) workflows.
  • IDE integration.

gnark exposes its APIs like any conventional cryptographic library. Complex solutions need API flexibility. For example gRPC and REST APIs, serialization protocols, monitoring, and logging can be easily added.

Example of how to prove knowledge of a pre-image

// Circuit defines a pre-image knowledge proof
// mimc(secret preImage) = public hash
type Circuit struct {
    PreImage frontend.Variable
    Hash     frontend.Variable `gnark:",public"`

// Define declares the circuit's constraints
func (circuit *Circuit) Define(api frontend.API) error {
    // hash function
    mimc, err := mimc.NewMiMC(api.Curve())

    // specify constraints
    // mimc(preImage) == hash
    api.AssertIsEqual(circuit.Hash, mimc.Hash(cs, circuit.PreImage))

    return nil
var mimcCircuit Circuit
r1cs, err := frontend.Compile(ecc.BN254, r1cs.NewBuilder, &mimcCircuit)
// witness 
assignment := &Circuit{
    Hash: "16130099170765464552823636852555369511329944820189892919423002775646948828469",
    PreImage: 35,
witness, _ := frontend.NewWitness(assignment, ecc.BN254)
publicWitness, _ := witness.Public()
pk, vk, err := groth16.Setup(r1cs)
proof, err := groth16.Prove(r1cs, pk, witness)
err := groth16.Verify(proof, vk, publicWitness)
assert := groth16.NewAssert(t)

var mimcCircuit Circuit

    assert.ProverFailed(&mimcCircuit, &Circuit{
        Hash: 42,
        PreImage: 42,

     assert.ProverSucceeded(&mimcCircuit, &Circuit{
        Hash: "16130099170765464552823636852555369511329944820189892919423002775646948828469",
        PreImage: 35,

gnark is fast


It is difficult to fairly and accurately compare benchmarks among libraries. Some implementations may excel in conditions where others may not. Results depend on target or available instruction set, CPUs and RAM.

Here we benchmark the same circuit using gnark, bellman (BLS12_381), and bellman_ce (BN254).


Number of constraints 100000 32000000 64000000
bellman_ce (s/op) 0.43 106 214.8
gnark (s/op) 0.12 27.1 53.9
Speed improvement x3.6 x3.9 x4.0

On large circuits, that’s over 1.18 million constraints per second.


Number of constraints 100000 32000000 64000000
bellman (s/op) 0.6 158 316.8
gnark (s/op) 0.19 41.4 80.6
Speed improvement x3.1 x3.8 x3.9


These benchmarks were executed on an AWS c5a.24xlarge instance, with hyper-threading disabled.

Results are not recent and will be updated.

Proving schemes and curves

Refer to the Proving schemes and curves section.

Questions or feedback? You can discuss issues and obtain free support on gnark discussions channel.
For paid professional support by Consensys, contact us at [email protected].